Covering Disruptive Technology Powering Business in The Digital Age

Home > Archives > News > ‘Liquid’ Machine-Learning System: A Revolutionary AI That Continuously Adapts
image
‘Liquid’ Machine-Learning System: A Revolutionary AI That Continuously Adapts
image
March 1, 2021 News

 

With all the promises of AI to industries, organisations and individuals, the technology still has a lot to live up to. One major worry about AI is that it lacks any ability of substantial decision-making a human can accomplish in varied and complex environments, say in life-or-death situations.

In an attempt to improve the capabilities of AI, developers use the patterns of neurons in animal brains to simulate its thinking capacity, which results in an artificial neural network. For so long, however, experts often modelled their AI algorithms to human brains, believing that complexity breeds more advanced the thought process. Based on a study conducted by researchers from the Massachusetts Institute of Technology (MIT), however, simplicity may have its benefits.

In their research, the MIT academics have designed a neural network that can adapt to the variability of real-world systems – drawing inspiration directly from a microscopic roundworm, the Caenorhabditis elegans.

“It only has 302 neurons in its nervous system, yet it can generate unexpectedly complex dynamics”, said Ramin Hasani, the study’s lead author. They have developed a type of neural network that not only learns during its training phase but also while doing its job – significantly improving its capability to adapt. This is different from most AI algorithms that are only trained before usage, which can limit its function over time.

The researchers dubbed these flexible algorithms as ‘liquid’ networks, as they can change their underlying equations to continuously adapt to new data inputs. This advancement in algorithms, according to the study, could aid decision-making based on data streams that change over time, including those involved in medical diagnosis and autonomous driving.

Hasani also added the fluidity of such ‘liquid’ networks makes it more resilient to unexpected or noisy data, like if heavy rain obscures the view of a camera on a self-driving car, so it is more robust. Essentially, this ‘liquid’ network can adapt to its environment and makes decisions depending on the changes it encounters.

Another important factor for having this type of network is the ability to process time-series data, which is important for today’s ever-changing digital landscape. According to Hasani, time-series data are both ubiquitous and vital to our understanding of the world.

“The real-world is all about sequences. Even our perception — you’re not perceiving images; you’re perceiving sequences of images. So, time-series data actually create our reality”, added Hasani. He also mentioned that video processing, financial data and medical diagnostic applications are examples of time-series central to society.

The ‘liquid’ network algorithm has edged out other state-of-the-art time-series algorithms by a few percentage points in accurately predicting future values in datasets, ranging from atmospheric chemistry to traffic patterns.

“In many applications, we see the performance is reliably high. Everyone talks about scaling up their network. We want to scale down, to have fewer but richer nodes”, said Hasani.

It is also in Hasani’s plans to keep improving the system and ready it for industrial application, believing that this kind of network could be an essential element of intelligence systems in the future.

(0)(0)